Some applications in video processing, e.g., edit propagation, may require explicit correspondence mapping between pixels in consecutive frames. Common approaches may be based on matching sparse feature points, dense optical flow estimations, and/or other techniques. However, finding a pixel-accurate mapping may inherently be an ill-posed problem, and existing dense approaches may require computationally expensive regularization and/or optimization.